etzion commited on
Commit
4124961
·
verified ·
1 Parent(s): 08b0e6a

org card: per-experiment hyperparameters and VQA/caption results

Browse files
Files changed (2) hide show
  1. README.md +27 -32
  2. index.html +17 -26
README.md CHANGED
@@ -341,38 +341,33 @@ generated at a 48-token budget, which clips verbose answers and penalises the
341
 
342
  ### Results
343
 
344
- Every model is scored on the same 1,991 judged items under the same rubric, so
345
- the column is comparable all the way down. **Read every gap against sigma**
346
- (see below): one checkpoint carries sigma 0.31, and the difference of two
347
- single checkpoints carries SE 0.44.
348
-
349
- | Model | VQA DM | VQA RSICD | CAP DM | CAP RSICD | MEAN4 |
350
- |---|---|---|---|---|---|
351
- | **Aspaklarya-101-step4500** | 74.16 | 70.56 | 82.53 | 79.64 | **76.72** |
352
- | **Aspaklarya-109-step3540** | 72.84 | 71.31 | 82.49 | 79.62 | **76.56** |
353
- | **Aspaklarya-109-step4500** | 73.1 | 70.21 | 82.95 | 79.25 | **76.38** |
354
- | **Aspaklarya-106-soup** | 74.57 | 69.38 | 82.38 | 79.1 | **76.36** |
355
- | **Aspaklarya-110-step3540** | 73.68 | 71.01 | 80.99 | 79.32 | **76.25** |
356
- | Aspaklarya-101-step3540 | 74 | 69.75 | 81.91 | 79.27 | **76.23** |
357
- | Aspaklarya-104-step2360 | 73.72 | 69.59 | 82.7 | 78.88 | **76.22** |
358
- | Aspaklarya-103-step2360 | 74.13 | 69.82 | 81.34 | 79.3 | **76.15** |
359
- | **Aspaklarya-110-step4500** | 74.03 | 69.8 | 81.6 | 78.94 | **76.09** |
360
- | Aspaklarya-108s2024-step2360 | 72.66 | 70.83 | 82.1 | 78.58 | **76.04** |
361
- | Aspaklarya-105-step1770 | 74.49 | 68.96 | 81.96 | 78.62 | **76.01** |
362
- | Aspaklarya-104-step1770 | 74.6 | 68.78 | 81.92 | 78.56 | **75.97** |
363
- | Aspaklarya-103-step1770 | 74.46 | 69.0 | 81.81 | 78.52 | **75.95** |
364
- | Aspaklarya-101-step2360 | 72.38 | 69.28 | 82.21 | 79.72 | **75.9** |
365
- | Aspaklarya-106b-soup | 72.77 | 69.99 | 81.53 | 78.73 | **75.75** |
366
- | Aspaklarya-102-step1770 | 73.45 | 68.93 | 81.56 | 78.6 | **75.64** |
367
- | Aspaklarya-102-step2360 | 73.24 | 68.71 | 82.0 | 78.49 | **75.61** |
368
- | Aspaklarya-105-step2360 | 73.18 | 68.7 | 82.09 | 78.41 | **75.59** |
369
- | Aspaklarya-108s2024-step1770 | 71.91 | 70.59 | 81.23 | 78.28 | **75.5** |
370
- | Aspaklarya-108s1234-step1770 | 72.24 | 69.38 | 81.06 | 78.25 | **75.23** |
371
- | Aspaklarya-108s1234-step2360 | 70.9 | 69.08 | 82.2 | 78.7 | **75.22** |
372
- | Aspaklarya-107-step1770 | 72.16 | 69.9 | 80.98 | 77.06 | **75.03** |
373
- | **Aspaklarya-020-step4500** | 66.18 | 70.65 | 82.79 | 79.99 | **74.9** |
374
- | in-house reference | 65.26 | 63.59 | 71.91 | 58.58 | **64.84** |
375
- | Qwen3-VL-8B (base) | 60.64 | 53.13 | 68.31 | 63.28 | **61.34** |
376
 
377
  ### Run-to-run sigma, measured
378
 
 
341
 
342
  ### Results
343
 
344
+ **One row per experiment: its best checkpoint.** Every model is scored on the
345
+ same 1,991 judged items under the same rubric, so the column is comparable all
346
+ the way down. The `ckpts scored` column says how many checkpoints that
347
+ experiment contributed; the best of n is an **order statistic**, so it is
348
+ biased upward by roughly 0.3-0.5 for n=4 and should not be read as that
349
+ recipe's expected quality. Use it to rank recipes, not to size effects.
350
+
351
+ For that, and for every individual checkpoint, see **Run-to-run sigma** below
352
+ and the full 36-row table in `rescore_slices.csv`. Seed replicates of one
353
+ recipe (108, 109, 110) are pooled into a single row here.
354
+
355
+ | Model | VQA DM | VQA RSICD | CAP DM | CAP RSICD | MEAN4 | ckpts scored |
356
+ |---|---|---|---|---|---|---|
357
+ | **Aspaklarya-101-step4500** | 74.16 | 70.56 | 82.53 | 79.64 | **76.72** | 4 |
358
+ | Aspaklarya-109-step3540 | 72.84 | 71.31 | 82.49 | 79.62 | **76.56** | 2 |
359
+ | **Aspaklarya-106-soup** | 74.57 | 69.38 | 82.38 | 79.1 | **76.36** | 1 |
360
+ | Aspaklarya-110-step3540 | 73.68 | 71.01 | 80.99 | 79.32 | **76.25** | 2 |
361
+ | **Aspaklarya-104-step2360** | 73.72 | 69.59 | 82.7 | 78.88 | **76.22** | 4 |
362
+ | **Aspaklarya-103-step2360** | 74.13 | 69.82 | 81.34 | 79.3 | **76.15** | 4 |
363
+ | Aspaklarya-108-step2360 | 72.66 | 70.83 | 82.1 | 78.58 | **76.04** | 4 |
364
+ | **Aspaklarya-105-step1770** | 74.49 | 68.96 | 81.96 | 78.62 | **76.01** | 4 |
365
+ | Aspaklarya-106b-soup | 72.77 | 69.99 | 81.53 | 78.73 | **75.75** | 1 |
366
+ | **Aspaklarya-102-step1770** | 73.45 | 68.93 | 81.56 | 78.6 | **75.64** | 4 |
367
+ | Aspaklarya-107-step1770 | 72.16 | 69.9 | 80.98 | 77.06 | **75.03** | 3 |
368
+ | **Aspaklarya-020-step4500** | 66.18 | 70.65 | 82.79 | 79.99 | **74.9** | 1 |
369
+ | in-house reference | 65.26 | 63.59 | 71.91 | 58.58 | **64.84** | - |
370
+ | Qwen3-VL-8B (base) | 60.64 | 53.13 | 68.31 | 63.28 | **61.34** | - |
 
 
 
 
 
371
 
372
  ### Run-to-run sigma, measured
373
 
index.html CHANGED
@@ -338,7 +338,8 @@ a { color: #0969da; }
338
  <p><code>MEAN4</code> is the unweighted mean, so a model cannot win by specialising on one slice. Judge: <code>gemma-4-31b-it-imiai</code>, greedy, temperature 0, scored 0-100.</p>
339
  <p>Two caveats on the shape of the benchmark. <strong>CAP DM carries 491 items, not 500</strong>, because the underlying 500-image sweep drew 491 DM / 509 RSICD. And <strong>RSICD VQA has no verbose rewrite</strong> - it is still the legacy question set generated at a 48-token budget, which clips verbose answers and penalises the 1XX models there.</p>
340
  <h3>Results</h3>
341
- <p>Every model is scored on the same 1,991 judged items under the same rubric, so the column is comparable all the way down. <strong>Read every gap against sigma</strong> (see below): one checkpoint carries sigma 0.31, and the difference of two single checkpoints carries SE 0.44.</p>
 
342
  <table style="border-collapse:collapse;margin:1rem 0;font-size:.85rem;border:1px solid #d0d7de"><thead><tr>
343
  <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">Model</th>
344
  <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">VQA DM</th>
@@ -346,32 +347,22 @@ a { color: #0969da; }
346
  <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">CAP DM</th>
347
  <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">CAP RSICD</th>
348
  <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">MEAN4</th>
 
349
  </tr></thead><tbody>
350
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-101-step4500</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">70.56</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.53</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.64</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.72</strong></td></tr>
351
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-109-step3540</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">72.84</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">71.31</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.49</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.62</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.56</strong></td></tr>
352
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-109-step4500</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.1</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">70.21</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.95</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.25</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.38</strong></td></tr>
353
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-106-soup</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.57</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.38</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.38</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.1</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.36</strong></td></tr>
354
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-110-step3540</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.68</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">71.01</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.99</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.32</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.25</strong></td></tr>
355
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-101-step3540</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.75</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.91</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.27</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.23</strong></td></tr>
356
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-104-step2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.72</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.59</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.7</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.88</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.22</strong></td></tr>
357
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-103-step2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.13</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.82</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.34</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.3</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.15</strong></td></tr>
358
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-110-step4500</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.03</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.8</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.6</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.94</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.09</strong></td></tr>
359
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-108s2024-step2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">72.66</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">70.83</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.1</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.58</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.04</strong></td></tr>
360
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-105-step1770</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.49</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.96</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.96</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.62</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.01</strong></td></tr>
361
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-104-step1770</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.6</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.78</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.92</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.56</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.97</strong></td></tr>
362
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-103-step1770</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.46</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.81</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.52</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.95</strong></td></tr>
363
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-101-step2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">72.38</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.28</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.21</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.72</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.9</strong></td></tr>
364
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-106b-soup</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">72.77</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.99</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.53</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.73</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.75</strong></td></tr>
365
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-102-step1770</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.45</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.93</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.56</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.6</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.64</strong></td></tr>
366
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-102-step2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.24</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.71</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.49</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.61</strong></td></tr>
367
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-105-step2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.18</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.7</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.09</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.41</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.59</strong></td></tr>
368
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-108s2024-step1770</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">71.91</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">70.59</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.23</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.28</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.5</strong></td></tr>
369
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-108s1234-step1770</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">72.24</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.38</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.06</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.25</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.23</strong></td></tr>
370
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-108s1234-step2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">70.9</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.08</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.2</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.7</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.22</strong></td></tr>
371
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-107-step1770</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">72.16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.9</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.98</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">77.06</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.03</strong></td></tr>
372
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-020-step4500</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">66.18</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">70.65</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.79</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.99</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>74.9</strong></td></tr>
373
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">in-house reference</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">65.26</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">63.59</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">71.91</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">58.58</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>64.84</strong></td></tr>
374
- <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Qwen3-VL-8B (base)</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">60.64</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">53.13</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.31</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">63.28</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>61.34</strong></td></tr>
375
  </tbody></table>
376
  <h3>Run-to-run sigma, measured</h3>
377
  <p>The previous version of this card said sigma had never been measured and that no one-point difference should be trusted. That has now been fixed, at the cost of four replication runs.</p>
 
338
  <p><code>MEAN4</code> is the unweighted mean, so a model cannot win by specialising on one slice. Judge: <code>gemma-4-31b-it-imiai</code>, greedy, temperature 0, scored 0-100.</p>
339
  <p>Two caveats on the shape of the benchmark. <strong>CAP DM carries 491 items, not 500</strong>, because the underlying 500-image sweep drew 491 DM / 509 RSICD. And <strong>RSICD VQA has no verbose rewrite</strong> - it is still the legacy question set generated at a 48-token budget, which clips verbose answers and penalises the 1XX models there.</p>
340
  <h3>Results</h3>
341
+ <p><strong>One row per experiment: its best checkpoint.</strong> Every model is scored on the same 1,991 judged items under the same rubric, so the column is comparable all the way down. The <code>ckpts scored</code> column says how many checkpoints that experiment contributed; the best of n is an <strong>order statistic</strong>, so it is biased upward by roughly 0.3-0.5 for n=4 and should not be read as that recipe&#x27;s expected quality. Use it to rank recipes, not to size effects.</p>
342
+ <p>For that, and for every individual checkpoint, see <strong>Run-to-run sigma</strong> below and the full 36-row table in <code>rescore_slices.csv</code>. Seed replicates of one recipe (108, 109, 110) are pooled into a single row here.</p>
343
  <table style="border-collapse:collapse;margin:1rem 0;font-size:.85rem;border:1px solid #d0d7de"><thead><tr>
344
  <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">Model</th>
345
  <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">VQA DM</th>
 
347
  <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">CAP DM</th>
348
  <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">CAP RSICD</th>
349
  <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">MEAN4</th>
350
+ <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">ckpts scored</th>
351
  </tr></thead><tbody>
352
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-101-step4500</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">70.56</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.53</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.64</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.72</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4</td></tr>
353
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-109-step3540</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">72.84</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">71.31</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.49</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.62</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.56</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2</td></tr>
354
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-106-soup</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.57</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.38</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.38</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.1</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.36</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1</td></tr>
355
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-110-step3540</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.68</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">71.01</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.99</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.32</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.25</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2</td></tr>
356
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-104-step2360</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.72</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.59</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.7</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.88</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.22</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4</td></tr>
357
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-103-step2360</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.13</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.82</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.34</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.3</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.15</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4</td></tr>
358
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-108-step2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">72.66</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">70.83</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.1</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.58</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.04</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4</td></tr>
359
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-105-step1770</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.49</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.96</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.96</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.62</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.01</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4</td></tr>
360
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-106b-soup</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">72.77</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.99</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.53</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.73</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.75</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1</td></tr>
361
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-102-step1770</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.45</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.93</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.56</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.6</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.64</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4</td></tr>
362
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-107-step1770</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">72.16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.9</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.98</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">77.06</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.03</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">3</td></tr>
363
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-020-step4500</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">66.18</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">70.65</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.79</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.99</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>74.9</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1</td></tr>
364
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">in-house reference</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">65.26</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">63.59</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">71.91</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">58.58</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>64.84</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">-</td></tr>
365
+ <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Qwen3-VL-8B (base)</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">60.64</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">53.13</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.31</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">63.28</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>61.34</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">-</td></tr>
 
 
 
 
 
 
 
 
 
 
 
366
  </tbody></table>
367
  <h3>Run-to-run sigma, measured</h3>
368
  <p>The previous version of this card said sigma had never been measured and that no one-point difference should be trusted. That has now been fixed, at the cost of four replication runs.</p>